func TestHandlerReconnects(t *testing.T) {
ctrl := gomock.NewController(t)
defer ctrl.Finish()
taskEngine := engine.NewMockTaskEngine(ctrl)
ecsclient := mock_api.NewMockECSClient(ctrl)
statemanager := statemanager.NewNoopStateManager()
closeWS := make(chan bool)
server, serverIn, requests, errs, err := startMockAcsServer(t, closeWS)
if err != nil {
t.Fatal(err)
}
go func() {
for {
select {
case <-requests:
case <-errs:
}
}
}()
ecsclient.EXPECT().DiscoverPollEndpoint("myArn").Return(server.URL, nil).Times(10)
taskEngine.EXPECT().Version().Return("Docker: 1.5.0", nil).AnyTimes()
ctx, cancel := context.WithCancel(context.Background())
ended := make(chan bool, 1)
go func() {
handler.StartSession(ctx, handler.StartSessionArguments{
ContainerInstanceArn: "myArn",
CredentialProvider: credentials.AnonymousCredentials,
Config: &config.Config{Cluster: "someCluster"},
TaskEngine: taskEngine,
ECSClient: ecsclient,
StateManager: statemanager,
AcceptInvalidCert: true,
})
// This should never return
ended <- true
}()
start := time.Now()
for i := 0; i < 10; i++ {
serverIn <- `{"type":"HeartbeatMessage","message":{"healthy":true}}`
closeWS <- true
}
if time.Since(start) > 2*time.Second {
t.Error("Test took longer than expected; backoff should not have occured for EOF")
}
select {
case <-ended:
t.Fatal("Should not have stopped session")
default:
}
cancel()
<-ended
}
func TestHandlerReconnects(t *testing.T) {
ctrl := gomock.NewController(t)
defer ctrl.Finish()
taskEngine := mock_engine.NewMockTaskEngine(ctrl)
ecsclient := mock_api.NewMockECSClient(ctrl)
statemanager := statemanager.NewNoopStateManager()
closeWS := make(chan bool)
server, serverIn, _, _, err := startMockAcsServer(t, closeWS)
if err != nil {
t.Fatal(err)
}
ecsclient.EXPECT().DiscoverPollEndpoint("myArn").Return(server.URL, nil).AnyTimes()
taskEngine.EXPECT().Version().Return("Docker: 1.5.0", nil).AnyTimes()
ended := make(chan bool, 1)
go func() {
handler.StartSession("myArn", credentials.NewCredentialProvider("", ""), &config.Config{Cluster: "someCluster"}, taskEngine, ecsclient, statemanager, true)
// This should never return
ended <- true
}()
start := time.Now()
for i := 0; i < 10; i++ {
serverIn <- `{"type":"HeartbeatMessage","message":{"healthy":true}}`
closeWS <- true
}
if time.Since(start) > 2*time.Second {
t.Error("Test took longer than expected; backoff should not have occured for EOF")
}
select {
case <-ended:
t.Fatal("Should never stop session")
default:
}
}
func _main() int {
defer log.Flush()
versionFlag := flag.Bool("version", false, "Print the agent version information and exit")
acceptInsecureCert := flag.Bool("k", false, "Do not verify ssl certs")
logLevel := flag.String("loglevel", "", "Loglevel: [<crit>|<error>|<warn>|<info>|<debug>]")
flag.Parse()
logger.SetLevel(*logLevel)
log.Infof("Starting Agent: %v", version.String())
log.Info("Loading configuration")
cfg, err := config.NewConfig()
// Load cfg before doing 'versionFlag' so that it has the DOCKER_HOST
// variable loaded if needed
if *versionFlag {
versionableEngine := engine.NewTaskEngine(cfg)
version.PrintVersion(versionableEngine)
return exitcodes.ExitSuccess
}
if err != nil {
log.Criticalf("Error loading config: %v", err)
// All required config values can be inferred from EC2 Metadata, so this error could be transient.
return exitcodes.ExitError
}
log.Debug("Loaded config: " + cfg.String())
var currentEc2InstanceID, containerInstanceArn string
var taskEngine engine.TaskEngine
if cfg.Checkpoint {
log.Info("Checkpointing is enabled. Attempting to load state")
var previousCluster, previousEc2InstanceID, previousContainerInstanceArn string
previousTaskEngine := engine.NewTaskEngine(cfg)
// previousState is used to verify that our current runtime configuration is
// compatible with our past configuration as reflected by our state-file
previousState, err := initializeStateManager(cfg, previousTaskEngine, &previousCluster, &previousContainerInstanceArn, &previousEc2InstanceID, acshandler.SequenceNumber)
if err != nil {
log.Criticalf("Error creating state manager: %v", err)
return exitcodes.ExitTerminal
}
err = previousState.Load()
if err != nil {
log.Criticalf("Error loading previously saved state: %v", err)
return exitcodes.ExitTerminal
}
if previousCluster != "" {
// TODO Handle default cluster in a sane and unified way across the codebase
configuredCluster := cfg.Cluster
if configuredCluster == "" {
log.Debug("Setting cluster to default; none configured")
configuredCluster = config.DEFAULT_CLUSTER_NAME
}
if previousCluster != configuredCluster {
log.Criticalf("Data mismatch; saved cluster '%v' does not match configured cluster '%v'. Perhaps you want to delete the configured checkpoint file?", previousCluster, configuredCluster)
return exitcodes.ExitTerminal
}
cfg.Cluster = previousCluster
log.Infof("Restored cluster '%v'", cfg.Cluster)
}
if instanceIdentityDoc, err := ec2.GetInstanceIdentityDocument(); err == nil {
currentEc2InstanceID = instanceIdentityDoc.InstanceId
} else {
log.Criticalf("Unable to access EC2 Metadata service to determine EC2 ID: %v", err)
}
if previousEc2InstanceID != "" && previousEc2InstanceID != currentEc2InstanceID {
log.Warnf("Data mismatch; saved InstanceID '%v' does not match current InstanceID '%v'. Overwriting old datafile", previousEc2InstanceID, currentEc2InstanceID)
// Reset taskEngine; all the other values are still default
taskEngine = engine.NewTaskEngine(cfg)
} else {
// Use the values we loaded if there's no issue
containerInstanceArn = previousContainerInstanceArn
taskEngine = previousTaskEngine
}
} else {
log.Info("Checkpointing not enabled; a new container instance will be created each time the agent is run")
taskEngine = engine.NewTaskEngine(cfg)
}
stateManager, err := initializeStateManager(cfg, taskEngine, &cfg.Cluster, &containerInstanceArn, ¤tEc2InstanceID, acshandler.SequenceNumber)
if err != nil {
log.Criticalf("Error creating state manager: %v", err)
return exitcodes.ExitTerminal
}
credentialProvider := auth.NewBasicAWSCredentialProvider()
awsCreds := auth.ToSDK(credentialProvider)
// Preflight request to make sure they're good
if preflightCreds, err := awsCreds.Credentials(); err != nil || preflightCreds.AccessKeyID == "" {
if preflightCreds != nil {
log.Warnf("Error getting valid credentials (AKID %v): %v", preflightCreds.AccessKeyID, err)
} else {
log.Warnf("Error getting preflight credentials: %v", err)
}
}
client := api.NewECSClient(awsCreds, cfg, *acceptInsecureCert)
if containerInstanceArn == "" {
log.Info("Registering Instance with ECS")
containerInstanceArn, err = client.RegisterContainerInstance()
if err != nil {
log.Errorf("Error registering: %v", err)
if retriable, ok := err.(utils.Retriable); ok && !retriable.Retry() {
return exitcodes.ExitTerminal
}
return exitcodes.ExitError
}
log.Infof("Registration completed successfully. I am running as '%v' in cluster '%v'", containerInstanceArn, cfg.Cluster)
// Save our shiny new containerInstanceArn
stateManager.Save()
} else {
log.Infof("Restored from checkpoint file. I am running as '%v' in cluster '%v'", containerInstanceArn, cfg.Cluster)
}
// Begin listening to the docker daemon and saving changes
taskEngine.SetSaver(stateManager)
taskEngine.MustInit()
go sighandlers.StartTerminationHandler(stateManager, taskEngine)
// Agent introspection api
go handlers.ServeHttp(&containerInstanceArn, taskEngine, cfg)
// Start sending events to the backend
go eventhandler.HandleEngineEvents(taskEngine, client, stateManager)
log.Info("Beginning Polling for updates")
err = acshandler.StartSession(containerInstanceArn, credentialProvider, cfg, taskEngine, client, stateManager, *acceptInsecureCert)
if err != nil {
log.Criticalf("Unretriable error starting communicating with ACS: %v", err)
return exitcodes.ExitTerminal
}
log.Critical("ACS Session handler should never exit")
return exitcodes.ExitError
}
func _main() int {
defer log.Flush()
flagset := flag.NewFlagSet("Amazon ECS Agent", flag.ContinueOnError)
versionFlag := flagset.Bool("version", false, "Print the agent version information and exit")
logLevel := flagset.String("loglevel", "", "Loglevel: [<crit>|<error>|<warn>|<info>|<debug>]")
acceptInsecureCert := flagset.Bool("k", false, "Disable SSL certificate verification. We do not recommend setting this option.")
licenseFlag := flagset.Bool("license", false, "Print the LICENSE and NOTICE files and exit")
blackholeEc2Metadata := flagset.Bool("blackhole-ec2-metadata", false, "Blackhole the EC2 Metadata requests. Setting this option can cause the ECS Agent to fail to work properly. We do not recommend setting this option")
err := flagset.Parse(os.Args[1:])
if err != nil {
return exitcodes.ExitTerminal
}
if *licenseFlag {
license := utils.NewLicenseProvider()
text, err := license.GetText()
if err != nil {
fmt.Fprintln(os.Stderr, err)
return exitcodes.ExitError
}
fmt.Println(text)
return exitcodes.ExitSuccess
}
logger.SetLevel(*logLevel)
ec2MetadataClient := ec2.DefaultClient
if *blackholeEc2Metadata {
ec2MetadataClient = ec2.NewBlackholeEC2MetadataClient()
}
log.Infof("Starting Agent: %s", version.String())
if *acceptInsecureCert {
log.Warn("SSL certificate verification disabled. This is not recommended.")
}
log.Info("Loading configuration")
cfg, cfgErr := config.NewConfig(ec2MetadataClient)
// Load cfg and create Docker client before doing 'versionFlag' so that it has the DOCKER_HOST variable loaded if needed
clientFactory := dockerclient.NewFactory(cfg.DockerEndpoint)
dockerClient, err := engine.NewDockerGoClient(clientFactory, *acceptInsecureCert, cfg)
if err != nil {
log.Criticalf("Error creating Docker client: %v", err)
return exitcodes.ExitError
}
ctx := context.Background()
// Create the DockerContainerChange event stream for tcs
containerChangeEventStream := eventstream.NewEventStream(ContainerChangeEventStream, ctx)
containerChangeEventStream.StartListening()
// Create credentials manager. This will be used by the task engine and
// the credentials handler
credentialsManager := credentials.NewManager()
// Create image manager. This will be used by the task engine for saving image states
state := dockerstate.NewDockerTaskEngineState()
imageManager := engine.NewImageManager(cfg, dockerClient, state)
if *versionFlag {
versionableEngine := engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
version.PrintVersion(versionableEngine)
return exitcodes.ExitSuccess
}
sighandlers.StartDebugHandler()
if cfgErr != nil {
log.Criticalf("Error loading config: %v", err)
// All required config values can be inferred from EC2 Metadata, so this error could be transient.
return exitcodes.ExitError
}
log.Debug("Loaded config: " + cfg.String())
var currentEc2InstanceID, containerInstanceArn string
var taskEngine engine.TaskEngine
if cfg.Checkpoint {
log.Info("Checkpointing is enabled. Attempting to load state")
var previousCluster, previousEc2InstanceID, previousContainerInstanceArn string
previousTaskEngine := engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
// previousState is used to verify that our current runtime configuration is
// compatible with our past configuration as reflected by our state-file
previousState, err := initializeStateManager(cfg, previousTaskEngine, &previousCluster, &previousContainerInstanceArn, &previousEc2InstanceID)
if err != nil {
log.Criticalf("Error creating state manager: %v", err)
return exitcodes.ExitTerminal
}
err = previousState.Load()
if err != nil {
log.Criticalf("Error loading previously saved state: %v", err)
return exitcodes.ExitTerminal
}
if previousCluster != "" {
// TODO Handle default cluster in a sane and unified way across the codebase
configuredCluster := cfg.Cluster
if configuredCluster == "" {
log.Debug("Setting cluster to default; none configured")
configuredCluster = config.DefaultClusterName
}
if previousCluster != configuredCluster {
log.Criticalf("Data mismatch; saved cluster '%v' does not match configured cluster '%v'. Perhaps you want to delete the configured checkpoint file?", previousCluster, configuredCluster)
return exitcodes.ExitTerminal
}
cfg.Cluster = previousCluster
log.Infof("Restored cluster '%v'", cfg.Cluster)
}
if instanceIdentityDoc, err := ec2MetadataClient.InstanceIdentityDocument(); err == nil {
currentEc2InstanceID = instanceIdentityDoc.InstanceId
} else {
log.Criticalf("Unable to access EC2 Metadata service to determine EC2 ID: %v", err)
}
if previousEc2InstanceID != "" && previousEc2InstanceID != currentEc2InstanceID {
log.Warnf("Data mismatch; saved InstanceID '%s' does not match current InstanceID '%s'. Overwriting old datafile", previousEc2InstanceID, currentEc2InstanceID)
// Reset taskEngine; all the other values are still default
taskEngine = engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
} else {
// Use the values we loaded if there's no issue
containerInstanceArn = previousContainerInstanceArn
taskEngine = previousTaskEngine
}
} else {
log.Info("Checkpointing not enabled; a new container instance will be created each time the agent is run")
taskEngine = engine.NewTaskEngine(cfg, dockerClient, credentialsManager, containerChangeEventStream, imageManager, state)
}
stateManager, err := initializeStateManager(cfg, taskEngine, &cfg.Cluster, &containerInstanceArn, ¤tEc2InstanceID)
if err != nil {
log.Criticalf("Error creating state manager: %v", err)
return exitcodes.ExitTerminal
}
capabilities := taskEngine.Capabilities()
// We instantiate our own credentialProvider for use in acs/tcs. This tries
// to mimic roughly the way it's instantiated by the SDK for a default
// session.
credentialProvider := defaults.CredChain(defaults.Config(), defaults.Handlers())
// Preflight request to make sure they're good
if preflightCreds, err := credentialProvider.Get(); err != nil || preflightCreds.AccessKeyID == "" {
log.Warnf("Error getting valid credentials (AKID %s): %v", preflightCreds.AccessKeyID, err)
}
client := api.NewECSClient(credentialProvider, cfg, httpclient.New(api.RoundtripTimeout, *acceptInsecureCert), ec2MetadataClient)
if containerInstanceArn == "" {
log.Info("Registering Instance with ECS")
containerInstanceArn, err = client.RegisterContainerInstance("", capabilities)
if err != nil {
log.Errorf("Error registering: %v", err)
if retriable, ok := err.(utils.Retriable); ok && !retriable.Retry() {
return exitcodes.ExitTerminal
}
return exitcodes.ExitError
}
log.Infof("Registration completed successfully. I am running as '%s' in cluster '%s'", containerInstanceArn, cfg.Cluster)
// Save our shiny new containerInstanceArn
stateManager.Save()
} else {
log.Infof("Restored from checkpoint file. I am running as '%s' in cluster '%s'", containerInstanceArn, cfg.Cluster)
_, err = client.RegisterContainerInstance(containerInstanceArn, capabilities)
if err != nil {
log.Errorf("Error re-registering: %v", err)
if awserr, ok := err.(awserr.Error); ok && api.IsInstanceTypeChangedError(awserr) {
log.Criticalf("The current instance type does not match the registered instance type. Please revert the instance type change, or alternatively launch a new instance. Error: %v", err)
return exitcodes.ExitTerminal
}
return exitcodes.ExitError
}
}
// Begin listening to the docker daemon and saving changes
taskEngine.SetSaver(stateManager)
imageManager.SetSaver(stateManager)
taskEngine.MustInit()
// start of the periodic image cleanup process
if !cfg.ImageCleanupDisabled {
go imageManager.StartImageCleanupProcess(ctx)
}
go sighandlers.StartTerminationHandler(stateManager, taskEngine)
// Agent introspection api
go handlers.ServeHttp(&containerInstanceArn, taskEngine, cfg)
// Start serving the endpoint to fetch IAM Role credentials
go credentialshandler.ServeHttp(credentialsManager, containerInstanceArn, cfg)
// Start sending events to the backend
go eventhandler.HandleEngineEvents(taskEngine, client, stateManager)
deregisterInstanceEventStream := eventstream.NewEventStream(DeregisterContainerInstanceEventStream, ctx)
deregisterInstanceEventStream.StartListening()
telemetrySessionParams := tcshandler.TelemetrySessionParams{
ContainerInstanceArn: containerInstanceArn,
CredentialProvider: credentialProvider,
Cfg: cfg,
DeregisterInstanceEventStream: deregisterInstanceEventStream,
ContainerChangeEventStream: containerChangeEventStream,
DockerClient: dockerClient,
AcceptInvalidCert: *acceptInsecureCert,
EcsClient: client,
TaskEngine: taskEngine,
}
// Start metrics session in a go routine
go tcshandler.StartMetricsSession(telemetrySessionParams)
log.Info("Beginning Polling for updates")
err = acshandler.StartSession(ctx, acshandler.StartSessionArguments{
AcceptInvalidCert: *acceptInsecureCert,
Config: cfg,
DeregisterInstanceEventStream: deregisterInstanceEventStream,
ContainerInstanceArn: containerInstanceArn,
CredentialProvider: credentialProvider,
ECSClient: client,
StateManager: stateManager,
TaskEngine: taskEngine,
CredentialsManager: credentialsManager,
})
if err != nil {
log.Criticalf("Unretriable error starting communicating with ACS: %v", err)
return exitcodes.ExitTerminal
}
log.Critical("ACS Session handler should never exit")
return exitcodes.ExitError
}
func TestHandlerDoesntLeakGouroutines(t *testing.T) {
ctrl := gomock.NewController(t)
defer ctrl.Finish()
taskEngine := engine.NewMockTaskEngine(ctrl)
ecsclient := mock_api.NewMockECSClient(ctrl)
statemanager := statemanager.NewNoopStateManager()
testTime := ttime.NewTestTime()
ttime.SetTime(testTime)
closeWS := make(chan bool)
server, serverIn, requests, errs, err := startMockAcsServer(t, closeWS)
if err != nil {
t.Fatal(err)
}
go func() {
for {
select {
case <-requests:
case <-errs:
}
}
}()
timesConnected := 0
ecsclient.EXPECT().DiscoverPollEndpoint("myArn").Return(server.URL, nil).AnyTimes().Do(func(_ interface{}) {
timesConnected++
})
taskEngine.EXPECT().Version().Return("Docker: 1.5.0", nil).AnyTimes()
taskEngine.EXPECT().AddTask(gomock.Any()).AnyTimes()
ctx, cancel := context.WithCancel(context.Background())
ended := make(chan bool, 1)
go func() {
handler.StartSession(ctx, handler.StartSessionArguments{"myArn", credentials.AnonymousCredentials, &config.Config{Cluster: "someCluster"}, taskEngine, ecsclient, statemanager, true})
ended <- true
}()
// Warm it up
serverIn <- `{"type":"HeartbeatMessage","message":{"healthy":true}}`
serverIn <- samplePayloadMessage
beforeGoroutines := runtime.NumGoroutine()
for i := 0; i < 100; i++ {
serverIn <- `{"type":"HeartbeatMessage","message":{"healthy":true}}`
serverIn <- samplePayloadMessage
closeWS <- true
}
cancel()
testTime.Cancel()
<-ended
afterGoroutines := runtime.NumGoroutine()
t.Logf("Gorutines after 1 and after 100 acs messages: %v and %v", beforeGoroutines, afterGoroutines)
if timesConnected < 50 {
t.Fatal("Expected times connected to be a large number, was ", timesConnected)
}
if afterGoroutines > beforeGoroutines+5 {
t.Error("Goroutine leak, oh no!")
pprof.Lookup("goroutine").WriteTo(os.Stdout, 1)
}
}
func TestHeartbeatOnlyWhenIdle(t *testing.T) {
testTime := ttime.NewTestTime()
ttime.SetTime(testTime)
ctrl := gomock.NewController(t)
defer ctrl.Finish()
taskEngine := engine.NewMockTaskEngine(ctrl)
ecsclient := mock_api.NewMockECSClient(ctrl)
statemanager := statemanager.NewNoopStateManager()
closeWS := make(chan bool)
server, serverIn, requestsChan, errChan, err := startMockAcsServer(t, closeWS)
defer close(serverIn)
go func() {
for {
<-requestsChan
}
}()
if err != nil {
t.Fatal(err)
}
// We're testing that it does not reconnect here; must be the case
ecsclient.EXPECT().DiscoverPollEndpoint("myArn").Return(server.URL, nil).Times(1)
taskEngine.EXPECT().Version().Return("Docker: 1.5.0", nil).AnyTimes()
ctx, cancel := context.WithCancel(context.Background())
ended := make(chan bool, 1)
go func() {
handler.StartSession(ctx, handler.StartSessionArguments{
ContainerInstanceArn: "myArn",
CredentialProvider: credentials.AnonymousCredentials,
Config: &config.Config{Cluster: "someCluster"},
TaskEngine: taskEngine,
ECSClient: ecsclient,
StateManager: statemanager,
AcceptInvalidCert: true,
})
ended <- true
}()
taskAdded := make(chan bool)
taskEngine.EXPECT().AddTask(gomock.Any()).Do(func(interface{}) {
taskAdded <- true
}).Times(10)
for i := 0; i < 10; i++ {
serverIn <- samplePayloadMessage
testTime.Warp(1 * time.Minute)
<-taskAdded
}
select {
case <-ended:
t.Fatal("Should not have stop session")
case err := <-errChan:
t.Fatal("Error should not have been returned from server", err)
default:
}
go server.Close()
cancel()
<-ended
}